The two greys were the same colour on the same day. They were sprayed eleven days apart, on a line whose mix ratio had been changed between shifts.

The situation

A European buyer ordered 420 upholstered dining chair frames in a powder-coated grey. The purchase order named the colour by reference to a physical chip from the previous production run, and added a single line: *colour to match approved sample, ΔE ≤ 1.5.*

The chairs were produced over eleven working days in a single order number. They were packed together, loaded into one container, and shipped. On arrival the buyer's own QA staff compared a sample from the front of the pallet against one from the back and saw a difference they described as *two batches of grey in one container.*

The difference was real. Measured properly it was a grey-scale step 3 — clearly visible side by side, and plainly unacceptable to a buyer who had approved one colour.

Nothing in the paperwork says there were two batches

The last row is the one that turns a fixable process variance into a commercial dispute. If the earlier units had been boxed separately and labelled with their production date, the buyer would have received two identified lots and could have sorted them in an afternoon. Packed together, they are one shipment of mixed shade with no way to separate them after unpacking.

Why a grey scale reading across batches is not the same test

Grey scale is an excellent instrument for one narrow job: comparing two samples viewed side by side under a defined illuminant, where the difference is small enough to read on a step scale. It is a poor instrument for the question the buyer was actually asking, which was *are all 420 frames the same colour as the approved chip.*

Two limitations did the damage here.

**Grey scale reads the pair you put in front of it, not the population.** A step 3 result means these two samples differ by three steps. It says nothing about how the other 418 frames are distributed. Reading a step 3 between two units and reporting *the batch shows a step 3* conflates a pairwise observation with a population measurement.

**Viewing conditions dominate the number.** A step 3 difference under a daylight-simulating D65 source at 1,000 lux can be a step 2 under D50 at 500 lux, and step 4 under a warm retail lamp. Any reading taken without a stated illuminant, lux level and observer geometry is not repeatable, and neither party can argue about it or reproduce it later.

The instrument is right. The method around it was missing.

How the sampling was done so the answer was usable

Run that way, the result separated cleanly. Lot one sat inside the tolerance against the approved chip. Lot two sat outside it, consistently and in the same direction — a shade slightly lighter and flatter across every reading point. **The two batches were not borderline. One passed and one failed.** That is a materially different conversation from *the batch has a colour issue*, and it is only available because the sampling separated the lots before measuring them.

• **Define the population before sampling.** The container was unpacked and the frames were sorted by packing date, because the packing date was the only surviving record that correlated with shade. That turned one shipment into two lots that could each be assessed. Sorting by nothing would have produced a random sample of a mixed population.
• **Draw a minimum sample from each lot, not from the shipment.** Five frames per lot, taken from front, middle and rear positions of the stacked bundles rather than from the top of each stack. Picking units that are visible on arrival is the standard way to get a flattering answer.
• **Compare against the archived reference, not against another unit.** The chip the buyer approved had been retained by the buyer's QC team and shipped back with the inspector. Every reading was made against that same physical reference, under the same lighting, in the same session. Comparing unit A to unit B produces a number that means nothing the next morning.
• **Record the instrument, the illuminant and the geometry.** Spectrophotometer make and model, D65 or D50 source, 10° standard observer, 45°/0° geometry, aperture size, and the number of reading points per panel. Without those six fields a ΔE is a private opinion.
• **Read more than one point per panel.** Colour varies across a coated surface. Taking a minimum, a maximum and a mean across the panel shows the spread, and a spread is often more informative than an average — a frame with a mean ΔE of 0.9 and a maximum of 2.1 is a different problem from a frame reading 0.9 everywhere.
• **State the tolerance as a limit with a method, not a number.** ΔE ≤ 1.5 is only actionable when paired with the illuminant it applies under and the sampling rule that produced it.

What the report said, and what it deliberately did not say

One sentence in that report did more work than the rest: *the two lots are separable by packing date, and the buyer can accept one and reject the other.* Without the sort the outcome would have been a claim against the whole container — a much larger sum, argued much more slowly, with a much lower chance of anyone agreeing.

• **The lot structure, established before measurement.** 420 frames, 211 from the first packing date and 209 from the second. Shade variation correlated with packing date at a level that left no realistic alternative explanation.
• **The finding as a population statement.** Lot 1: within tolerance. Lot 2: outside tolerance on ΔE, with the mean and range for each. Not *the shipment shows a colour difference* — that phrasing puts the verdict on goods that are half compliant.
• **The measurement conditions in full**, as listed above, so the buyer can reproduce the reading or dispute it on the record.
• **The finding about process control, not about the operator.** The colour reference was unchanged, the powder batch was unchanged, and the mix ratio was not recorded anywhere. The report states that the process was unable to *demonstrate* consistency, which is a different and more accurate claim than stating that the goods were made inconsistently.
• **Explicitly no claim about intent or about falsification.** Nothing in the evidence supported either, and a report that reaches for them spends the credibility it needs for the parts it got right.

What to put in the specification

Colour is one of the most frequently written lines in a furniture purchase order and one of the least frequently made checkable. Six additions turn it into something an inspector can either verify or fail to verify:

• **Name the measurement method, not just the number.** *ΔE (CIE76, D65, 10° observer) ≤ 1.5.* CIE76 and CIE94 and CIE2000 produce different numbers from the same pair of samples, and a tolerance without a method is a preference rather than a requirement.
• **State the reference as a physical control sample.** *Match to approved control sample held at the factory and available at inspection.* A control sample that lives in the sales office and travels between three sites is not a reference; it is a rumour.
• **Require the mix to be recorded per shift.** *Mix ratio, booth pressure and cure time recorded for each production shift and retained for twelve months.* This single clause is the difference between a process that can be audited and one that cannot.
• **Set a lot definition.** *One production date, one shift team or one recorded set of process parameters = one lot; lots packed and labelled separately.* This is what would have made the shipment separable on arrival.
• **Specify the sampling plan.** *Five units per lot from front, middle and rear; minimum, maximum and mean ΔE recorded per unit against the control sample.* A tolerance with no sample size gets sampled by whoever is nearest the door.
• **Define the outcome per lot, not per shipment.** *A lot outside tolerance is rejected or reworked; other lots in the same shipment remain eligible for acceptance.* This clause is the difference between a correction and a total loss, and it is far cheaper to write than to negotiate.